Load cell zero offset - how much is too much?

[magic]

Is anyone familiar with typical specifications and behavior of load cells?

I'm measuring the load cell of a digital scale which struggles with repeatability; it's some no-name thing only marked with capacity and "class C3".
I found that excitation is a stable 5V, zero load output is -13mV (2.6mV/V) and full load sensitivity is 7mV (1.4mV/V).

I have no datasheet of course, but I tried looking up specs of similar load cells. As far as I see, my measured sensitivity figure is fairly average, but zero balance is usually quoted in units of "%FS" with a single-digit number, so I presume offset should be at least an order of magnitude lower than full scale output, right?

So this cell is dead for good, isn't it?

[David Hess]

Good load cells are trimmed during manufacturing to have a small zero offset, so if you find one which is far off of zero, it implies that it was overloaded and damaged.

[SiliconWizard]

Yep an excessive zero offset is a sign of damage.
I'm assuming you measured the differential output, ideally disconnected from the differential amplifier (there must be one), if possible, just to rule out a fault in the input stage of the amplifier itself rather than the load cell.

[Kleinstein]

Normally the offset should be smaller. For a low cost product it is still possible to have a load cell with quite some offset. Modern electronics has less of a problem with compensating for offset and loosing 1-2 bit is not a big issue.

[David Hess]

Normally the offset should be smaller. For a low cost product it is still possible to have a load cell with quite some offset. Modern electronics has less of a problem with compensating for offset and loosing 1-2 bit is not a big issue.

In the past offset was a problem because it is difficult to remove without introducing drift.

[magic]

So the conclusion is that it could be broken, cheap or both? Well, it doesn't look expensive, that much I can tell, so I will have to investigate further.

I'm assuming you measured the differential output, ideally disconnected from the differential amplifier (there must be one), if possible, just to rule out a fault in the input stage of the amplifier itself rather than the load cell.

Yes, differential voltage between green and white wire. They go straight to the ADC chip, so there isn't much electronics in there. Good point about desoldering the wires, I haven't tried yet because more disassembly is required first, I just probed the signals on the board.

What I have tried, though, is shorting the two traces together. Initially it looked OK (zero reading, no reaction to pressing on the load cell), but the reading slowly drifts in the opposite direction as previously. I initially thought it's perhaps thermal EMF, but left overning it reached 2% of full scale and still keeps growing

My short circuit still measures 2Ω including multimeter probes and reading on the scale varies 10% in response to multimeter test current. Not sure what's happening, maybe some digital drift compensation gone wrong

[RoGeorge]

I'm measuring the load cell of a digital scale which struggles with repeatability

AFAIK usual consumer grade scale is made with piezo elements (not resistive load cells).  For piezo, one has to measure the total charge generated by all the cells (charge Q is proportional with the displacement IIRC), which is proportional with the applied weight.  My guess is they measure integral of generated voltage when a weight is placed on the scale.  These are personal expectations, I didn't check how it is in practice.

What I know for sure, and this one I've checked, recent models will lie about the exact weight (because it is not possible to get a repeatable value).  If the difference from the last measurement isn't too high, it will keep showing the previous number.  My scale will keep showing the same number if I step on it with, or without a 0.5L bottle of water in hand.

If wait a few minutes and try again, it will update the number and the water bottle will be included in the displayed number.

[magic]

Piezo would be some open circuit, high impedance thing, wouldn't it?
This load cell is powered with 5V and outputs two 2.5V signals which shift apart in response to loading the cell.
In-circuit and powered down, I measured ~400Ω between the two outputs with a DMM.
Looks like a resistive Wheatstone bridge to me.

Consumer scales contain all sorts of fascinating software features
https://www.eevblog.com/forum/dodgy-technology/i-cracked-the-mystery-of-chinese-precision-jewelry-scales/

[Kleinstein]

I don't think there are normal scales that use piezoelectric sensors, at last not in the direct (DC) mode.  Some cheap scales use capacitive read_out to measure the deflection, but that is inherently with some AC driven bridge.

A resistance of some 400 ohm points clearly to DMS and so does a sensitivity of some 2 mV/V. The standard value for DMS is 350 ohm and the sensitivty is perfectly in that range.

The drift seen with a shorted sensor points to a problem with the amplifier / ADC.
It is possible to build the amplifier so that the bridge resistance effect the gain - so the shorted case may also increase the gain and thus make drift look more dramatic than it actually is.

[David Hess]

AFAIK usual consumer grade scale is made with piezo elements (not resistive load cells).  For piezo, one has to measure the total charge generated by all the cells (charge Q is proportional with the displacement IIRC), which is proportional with the applied weight.  My guess is they measure integral of generated voltage when a weight is placed on the scale.  These are personal expectations, I didn't check how it is in practice.

You might be confusing semiconductor strain gauges with piezo elements.  The former have much higher gauge factors than metal film strain gauges making them much more sensitive so signal conditioning is easier, but are not as accurate.

[RoGeorge]

Seems I was wrong, thank you all for clarifying.

Don't know where from, or why did I believe the sensor was a piezo.  Even the part with repeating the last number instead of the actual weight seems weird.  Might be something else, because I've just double checked moments ago with various weights, and it tends to repeat the last value, just that it "prefers" certain values (in steps of 0.4kg - the scale has 0.1kg resolution), unless I wait a few minutes and try again, then it prefers other values, no idea why so.

I'll strikeout my previous post, shouldn't have written that at all, sorry.

[magic]

Might be something else, because I've just double checked moments ago with various weights, and it tends to repeat the last value, just that it "prefers" certain values (in steps of 0.4kg - the scale has 0.1kg resolution), unless I wait a few minutes and try again, then it prefers other values, no idea why so.

When I played with the shitty scale which cheats reviews by rounding to full grams, I found similar "preference" for discrete steps of 3 counts. Dividing its full capacity (200g) by "step size" (3mg) the result is close to 65536, which I consider possibly not a coincidence.

It's a tinfoil hat theory because I haven't disassembled that scale to see what ADC was inside, but I suspect that it could be a 16 bit ADC. Still no sure explanation why it sometimes changes the "preference", but maybe there is digital drift compensation working under the assumption that if the reading creeps slowly and steadily over time, it's load cell drift rather than the weighed mass actually changing. Such a digital mechanism could work with finer resolution than the ADC by means of long term accumulation and averaging, explaining the occasional shifts of "preferred" numbers.

Similarly, 0.4kg · 256 is 100kg, so what's the maximum capacity of your scale?

edit
There is the issue mentioned by Kleinstein of ADC bits being "lost" when full scale load cell output is less than full scale input to the ADC, particularly likely to occur in dodgy and not-so-optimized designs. So it's not as simple as checking the number of bits of the ADC - one needs to consider what is the actual output of the load cell, any amplifiers (if present), and the internal gain of the ADC.

[RoGeorge]

Could be, on the glass plate it says "Max 150kg d=0.1kg   Max 330lb d=0.2lb".

Never mind, did you find any clues on yours, is the measurement uncertainty coming from the sensors, or from the ADC?

[r6502]

Hello all,

There is a 2nd indicator that a load cell is defect: defect load cells tent to drift.

When you apply a load and release it, the output signal should return to the zero signal +- 1 digit of its resolution within a defined time. This value is given as  "Zero Return (creep) in 2 minutes" in "±% of Applied Load" with this value, you cold calculate the resolution, when you use the max load, and it is typically 3000d = 0,033%

If the output signal dos not return to this point it may have been overloaded in the past.

Example: when you have a load cell for max 3000g, it should return to 0 + - 1g after a time of 2min, and the zero point should be stable within + - 1g if not it is damaged.

BR Guido

[magic]

Never mind, did you find any clues on yours, is the measurement uncertainty coming from the sensors, or from the ADC?

What uncertainty and which scale?

Either way, the answer is probably no. The cheating scale I have returned to the seller after a few days without looking inside because I wasn't sure how to take it apart without damage. All I know about its behavior is documented in the old thread. The broken one isn't mine so I'm not familiar with it at all, and in its current state it drifts even with shorted ADC input and despite there being no detectable voltage across the short circuit (less than a few μV, certainly not a few % of full scale load cell output which was found to be several mV).

I'm not sure what's going on, there is a dumb off-the-shelf ADC chip and an unknown other IC, the ADC should simply be reporting constant readings and the controller ought to interpret them as "nothing changes". I suppose I could verify the ADC by scoping serial communications and maybe attach a load cell simulator made of resistors, recalibrate and see if the drift persists. But given that the load cell is dubious too I'm not sure if I want to bother debugging those electronics.

[wraper]

1.4 mV/V is an oddball value, normally they are either 1 mV/V or 2 mV/V. Zero offset super way too high for a good load cell. You can expect a few % of full scale offset from a cheap load cell but not something like this.
EDIT: This is of course if it's single load cell with a bridge, not 4 separate load cells connected together where expected offset is much higher, especially if load cells are not well matched.

[r6502]

Hello all,

I just looked at the loadcells, I have in stock, it is a Telda model 1004 with a max load of 300g, and measured the offset voltage in 2 conditions:

It has an offset in the range of 20µV to 30µV, 3 different load cells tested. This works only, if the point of load is free, and the point of attachement is fixed. In my test setup there was a groove of about 0.5mm milled away that the load cell could free move, when a load was applied. refer to "loadcell_1.jpg"

When you put the loadcell just on the desk, like shown in the photo "loadcell_2.jpg" ther is a large offset of  about -500µV.

This loadcell, model 1004 is very sensitive with a max load of 300g and a short term resolution of  30000d, so for loacells, that are specified for 3000d only this may not be that critical. But try to fix the loadcell at the fixing point, make sure that the the point where the load is applied, could free move and then observe the offset and it's drift. Observe also the drift when you apply a constant load to the cell.

Guido